Presentation 2020-03-01
Online Learning for A Repeated Markovian Game with 2 States
Shangtong Wang, Shuji Kijima,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We consider a new problem of learning in repeated games. In our model, the players play on one of the several game matrices $M_1,M_2,... in mathcal{M}$ in each round. The set of matrices $mathcal{M}$ is fixed in each sequence of games. The matrix to use in the next round depends on the matrix is being used and the decision of players in the current round. We regard the row player as the learner and his goal is to minimize his loss in the sequence of games. In this paper, we particularly concerned with a simple instance with only 2 matrices. We show that directly applying the existing multiplicative weights algorithms or "Follow the perturbed leader" algorithm cannot achieve $o(T)$ regret, where $T$ is the length of the sequence of the games.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Online LearningRepeated GameAlgorithm
Paper # COMP2019-55
Date of Issue 2020-02-23 (COMP)

Conference Information
Committee COMP
Conference Date 2020/3/1(1days)
Place (in Japanese) (See Japanese page)
Place (in English) The University of Electro-Communications
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Toshihiro Fujito(Toyohashi Univ. of Tech.)
Vice Chair Shinichi Nakano(Gunma Univ.)
Secretary Shinichi Nakano(Kumamoto Univ)
Assistant Kazuhisa Seto(Seikei Univ.)

Paper Information
Registration To Technical Committee on Theoretical Foundations of Computing
Language ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Online Learning for A Repeated Markovian Game with 2 States
Sub Title (in English)
Keyword(1) Online LearningRepeated GameAlgorithm
1st Author's Name Shangtong Wang
1st Author's Affiliation Kyushu University(Kyushu Univ.)
2nd Author's Name Shuji Kijima
2nd Author's Affiliation Kyushu University(Kyushu Univ.)
Date 2020-03-01
Paper # COMP2019-55
Volume (vol) vol.119
Number (no) COMP-433
Page pp.pp.65-68(COMP),
#Pages 4
Date of Issue 2020-02-23 (COMP)